Design and Implementation of a Low Cost, Hierarchical and Modular Avionics Architecture for the DragonFly UAVs

Jung Soon Jang, Claire J. TomlinAIAA 2002 · DOI

TL;DR

This paper presents a low-cost, hierarchical, and modular avionics architecture for the Stanford DragonFly UAV, built on the QNX Neutrino real-time operating system. The design emphasizes isolating user applications from critical system services using a client-server model with shared memory for efficient inter-process communication, aiming for robust and adaptable UAV control.

What Problem Does This Paper Address?

The paper tackles the challenge of designing safety-critical avionics for Unmanned Aerial Vehicles (UAVs), specifically the Stanford DragonFly. Traditional federated architectures, while offering fault isolation, are resource-intensive. The authors aim to develop an integrated modular avionics (IMA) approach that leverages Commercial Off-The-Shelf (COTS) components for cost-effectiveness and ease of modification, while still ensuring reliability and performance. A key problem is the integration of control engineering and software engineering, which often remain separate disciplines, leading to communication gaps and suboptimal designs. This paper seeks to bridge this gap by presenting an architecture that facilitates the development and testing of control algorithms within a robust software framework.

Methods

The core of the proposed solution is a hierarchical and modular software architecture based on the Client-Server model and the QNX Neutrino real-time operating system.

  • Client-Server Architecture: This design separates user-defined applications (clients) from safety-critical system services and hardware management (servers). This isolation simplifies application development, allows for independent testing, and enhances system robustness by running services in their own virtual memory address spaces.
  • QNX Neutrino RTOS: Chosen for its microkernel design, real-time capabilities, and support for client-server interactions.
  • Shared Memory for IPC: To overcome the bandwidth limitations of traditional message passing in embedded systems, the authors employ shared memory for inter-process communication (IPC). This allows clients and servers to directly read and write data, significantly increasing throughput.
  • Mutexes for Synchronization: To manage concurrent access to shared memory and prevent data inconsistency, mutexes are used to ensure mutual exclusion, with a priority-based mechanism to grant access to the highest-priority waiting thread.
  • Prioritized Round-Robin (RR) Scheduling Algorithm: A custom scheduling algorithm is implemented to manage CPU time among various threads. It prioritizes higher-priority threads, interrupting lower-priority ones if necessary, while also ensuring fair sharing among threads of the same priority through RR.
  • Integrated Modular Hardware Architecture: The hardware is built using PC/104 stackable modules and standard RS232 interfaces, enabling the use of COTS components and facilitating expansion and modification. The hardware is functionally separated into Computing, Actuating, and Gathering modules.
  • Open Control Platform (OCP) Integration: The avionics architecture is developed in parallel with Boeing's OCP, a software infrastructure for developing and testing control algorithms. The OCP is used for ground station functionalities, demonstrating interoperability.

Datasets

Not discussed in the paper. The paper focuses on the design and implementation of the avionics system and its testing through car tests and hardware-in-the-loop simulations, rather than on training or evaluating against specific datasets.

Key Insights and Ideas

  • Bridging Control and Software Engineering: The paper highlights the critical need for tighter integration between control engineers and software engineers. By designing an architecture that abstracts low-level OS details and provides a clear interface for control applications, it empowers control designers to focus on algorithms rather than embedded coding complexities.
  • Shared Memory as a High-Bandwidth IPC Solution: For embedded systems with high data throughput requirements, shared memory, despite its synchronization challenges, offers a significant performance advantage over traditional message passing. The use of mutexes effectively mitigates the risks of data inconsistency.
  • Hierarchical Design for Robustness and Maintainability: The client-server and layered architecture (Services, Management, Client) provides a clear separation of concerns, making the system more robust, easier to debug, and more adaptable to future hardware or software changes.
  • COTS for Cost-Effectiveness: The reliance on PC/104 and standard interfaces demonstrates a practical approach to building capable avionics systems at a lower cost, making advanced UAV technology more accessible.
  • Prioritized Scheduling for Real-Time Guarantees: The custom prioritized RR scheduling algorithm is crucial for ensuring that time-critical tasks meet their deadlines, a fundamental requirement for real-time control systems.

Key Results

  • Successful Implementation: The paper details the successful design and implementation of the DragonFly avionics system, integrating hardware and software components.
  • Car Test Validation: Preliminary tests involving driving the avionics system in a car around a campus loop demonstrated that all system activities, including scheduling, inter-process communication, and resource management, functioned seamlessly.
  • Hardware-in-the-Loop (HIL) Simulation: A HIL simulation testbed was developed, running on both Windows and QNX, to validate control algorithms and hardware. This setup allows for realistic testing without actual flight.
  • Performance Metrics: Figures show execution intervals for "GPS" and "DAQ" threads, indicating system scheduling capabilities. For example, the worst-case interval for the "DAQ" thread was 30 msec, and for "GPS" was 1.22 sec, with no lost interrupts observed for a significant number of services.
  • Jitter Analysis: The paper discusses observed jitter in thread execution intervals, particularly for the "GPS" thread, and proposes potential solutions like upgrading the GPS receiver or using timer-based polling instead of interrupts.

Limitations and Future Work

  • GPS Jitter: The paper acknowledges jitter in the "GPS" thread's execution interval, even with high priority. Future work may involve upgrading the GPS receiver or modifying the GPS code structure to poll data using timers, though this comes at the cost of increased CPU usage.
  • Hardware Redundancy: Due to strict payload, power, and budget limitations, hardware redundancy (duplex or triplex) was not implemented. Instead, the system relies on robust software and hardware partitioning for fault tolerance.
  • Flight Tests: While car tests and HIL simulations were performed, the paper mentions plans for flight tests of the single-vehicle control and navigation, and dual-vehicle algorithms on the DragonFly UAVs in the near future.
  • OCP Integration Limitations: While OCP offers a platform-independent interface, low-level services not supported by its core components (like device drivers and interrupt management) still require manual implementation, similar to the "services" in the DragonFly architecture.
  • Future Modifications: The modular design is intended to ease future modifications and growth, allowing for the addition of new modules or functions by simply adding new software partitions.

Summary updated

Abstract

We present the design and implementation of the safety critical avionics for the Stanford DragonFly Unmanned Aerial Vehicles (UAVs). The software architecture of the avionics is based on the Server-Client architecture of QNX Neutrino, the real-time operating system used. Our architecture is hierarchical and modular: it isolates user-defined applications from underlying low-level system services for implementing inter-process communication, data-acquisition, and associated hardware management. Also, we employ a new run-time scheduling algorithm to maximize the execution of tasks within a given deadline. The integrated hardware architecture is based on standard PC/104 and RS232 technology, making it possible to use Commercial O-The-Shelf (COTS) components and provide an ecient means of communication between components. We have developed this architecture in parallel (and in conjunction) with Boeing’s Open Control Platform (OCP) architecture, which is a new software infrastructure based on real-time CORBA technology, and we use the OCP to implement ground station functionalities. We present the design principles and choices, the resulting avionics architecture, and the implementation in a robust and compact avionics package. We then present initial results of the avionics in car tests, and discuss the results.
BibTeX
@inproceedings{jang2002design,
  author = {Jung Soon Jang and Claire J. Tomlin},
  title = {{Design and Implementation of a Low Cost, Hierarchical and Modular Avionics Architecture for the DragonFly UAVs}},
  booktitle = {AIAA},
  year = {2002},
  doi = {10.2514/6.2002-4465},
}
Figure